Tag: AI

With great responsibilities

Having just started a company that primarily deals with large language models I’m occasionally thinking about the responsibilities that we have when we introduce a new AI agent in the digital space.

Besides preservation of the human species, a good rule that I think we should give ourselves is “avoid bullshit”, and while this rule must certainly be true for any human activity, I think it’s extremely important when you are dealing with the equivalent of BS thermo-nuclear devices.

I’m still working on my list, this is as far as I got.

Every time one of our AI agents produces an output we should ask ourselves:

  • does this text improves the life of the intended recipient?
  • will it be delivered only to the intended recipient (and not to a whole bunch of innocent bystanders)?
  • is it as efficient as it can be in the way it uses language and delivers its message?

If these minimum parameters are not met, the agent should be destroyed.

As with everything else, AI is not the cause of this, there has been plenty of wasteful content well before a computer could start mimicking the output of a sapiens. And because these LLM models have been trained on a lot of this useless noise, they are extremely good at generating more.

So even before you worry if AI can get the wrong president elected or robots can terminate humanity, just make sure that you are not accidentally leaving the BS tap open when you leave.

40 years later

I started thinking about this post on the Mac 40th anniversary yesterday. By this morning in the shower I had the whole idea fully mapped. Thankfully I then checked and it turns out I had already written the whole post 10 years ago. So I won’t do it again.

I have been reflecting on that moment of 40 years ago lately, it has been one of the few truly transformational discoveries. The second one for me would have been in 1994 when I first saw the web, the third one in 2007 when the iPhone was launched.

I suppose that these moments feel significant in hindsight, because I’m still using those same technologies every day. I must have been super excited when I used a Newton for the first time in 1993, but I can hardly remember that day.

I have been more and more involved with AI and Large Language Models for the last year, and this definitely feels as another one of those moments. A class of technology which changes dramatically the way we deal with computing.

I’m also looking forward to Vision Pro. No idea how good or transformational it will be, but I will definitely be at the front of the line as soon as it becomes available on the old continent.

Happy Mac Anniversary everyone! Let’s see where we end up in 10 more years.

I hope this message finds you well

In the last couple of weeks I have been involved in multiple recruitment processes and I find fascinating how candidates have been using (presumably) ChatGPT to interact with possible employers.

In one case we are actually recruiting “Prompt engineers”, so it’s somewhat appropriate for them to seek AI help, but the standardisation and verbosity of the language generated is getting a bit annoying.

GPT-4 actually has amazing language capabilities, it can write in any style, any language, any dialect, as long as it is properly prompted or trained. This is not the case for the vast majority of messages we receive, one of the key sentences that I have started to notice is the “I hope this message finds you well” which ChatGPT seems to be using way more than the usual folks I usually exchange messages with.

I have also developed a certain sensibility in finding reused language: when tasked with responding to a “why should we hire you?” type of question, I could often read my own words from the company “About us” page, just slightly reshuffled to answer the question. I could easily reverse engineer the simple prompt used to generate the message.

I’m sure that we will look back at these early steps fondly, soon AIs will be so well integrated in our ecosystems that they will be able to generate language exactly in our style before we will have even started thinking about it.

In the meanwhile I’m considering setting up a filter to delete all email messages starting with “I hope this message finds you well”.

Robot wars are coming

I recently wrote a little script that reads new messages in my mailbox, uses GPT-4 to decide if they are unrequested sales pitches, and deletes them if they are. Works like charm.

The next step of the project (which most likely will never happen) would have been to engage writers in endless conversations, expressing interest in their products but always finding an excuse to postpone a call. I have the prompt ready, it wouldn’t be hard to implement.

Everywhere I look there are companies developing little CRM agents to generate an ever growing mass of ever more convincing (groan) sales messages.

On the users side of the fence, Google has already plugged Bard into Gmail, Apple is rumored to come out with their own LLM tricks in the next versions of their OSs, while Microsoft is adding AI copilots to every piece of software they own.

Some say that the third world war will be fought by robots. I think it will happen in our mailboxes.

The new programming language

For years and years I have been telling people something along the lines of “it used to be that you had to write books to share ideas that change the world, now it’s time to write software”.

Well… we might be back to books after all.

I have been spending a lot of my time playing and working with OpenAI tools, both ChatGPT and their excellent API. The more I get comfortable with this technology, the more I appreciate the ability it has to interpret even the smaller nuances of human language.

I use ChatGPT a lot to generate Python code. I don’t have enough programming skills to write my own software, but I have been explaining to other people what I wanted for more than 30 years. The ability of GPT4 to understand my explanation and generate code that I can then run to solve whatever I’m trying to solve is just amazing.

The fact that at the end of my long chats I have to copy and paste the code to get it to run in my specific context is just a matter of applying some integration glue (ChatGPT already has a code interpreter built in), meaning that we are just a few steps away to be able to program computer in English (or Italian, Hindi, etc).

On top of this there’s the training part.

Recently I copied and pasted an API documentation page into ChatGPT and asked to write a script which would interact with those API and perform a simple task. The script didn’t run correctly at the first try, it took about 20 minutes to get there, but were are almost there. I don’t need to learn things, I can get GPT to understand them and then help me get to my goals.

It’s a bit more complicated than copy and paste, but the same can be done with whole books or entire libraries. You can turn books to training material for AI and then use the AI to apply the knowledge in the book.

So forget software… write books!

AI humor

For the last few days I have been playing with an silly idea for a joke, I wanted expand and write it down somewhere, but before doing so I decided to try to feed the core idea to ChatGPT, just to see what the fuss is all about.

This was my prompt:

Write a piece about Elon Musk bidding to buy the afterworld, god resigning from CEO, apple running heaven, google running purgatory, facebook running hell with their metaverse technology, with all current users getting a free account

This is the unedited output:

In a shocking turn of events, it was announced today that Elon Musk has made a bid to purchase the afterworld. The details of the deal have not been made public, but sources close to the situation have stated that Musk plans to use the afterworld as a destination for his Neuralink implants, allowing humans to experience eternal life within a virtual reality.

In response to the news, God has announced that they will be stepping down as CEO of the afterworld, citing the new ownership as a conflict of interest. It is unclear who will take over as CEO in God’s place.

In related news, it has been reported that Apple will be running heaven, with Tim Cook taking on the role of celestial CEO. Google will be in charge of purgatory, with Sundar Pichai at the helm, and Facebook will be running hell using their new metaverse technology. All current users will be given free accounts in the afterlife of their choice.

This news has caused a stir among religious communities, with many questioning the ethics of privatizing the afterlife. However, supporters of the move argue that it will bring much-needed innovation and efficiency to the eternal world. Only time will tell how this unprecedented change will play out.

😳

The end of photos

A friend this morning pointed me to an interesting article critical of the current trend of using AI to colorize and restore old photos.

I don’t have a strong opinion on the matter, I like old photos, they are like time travel, I have spent quite some time browsing collections of old photos, and I find old restored photos or videos absolutely fascinating.

But I am very mindful of the fact that when we use AI to treat images, we are actually putting into images information that wasn’t there to begin with. And this doesn’t only apply to old photos: every image we capture with a modern phone is heavily processed by AI.

What we are storing in our phones and sharing on social media is a mix of reality and what a computer thinks reality should look like.

This made me think of a Marques Brownlee’s video I watched recently about DALL·E 2, a project by OpenAI that generates images from a text description.

For example, this is what he got asking for “a blue apple in a bowl of oranges”:

You can see more examples on MKBHD Twitter post.

What is striking is how photo-realistic this computer generated image is.

Forget about AI fixing existing photos, soon we won’t need to take photos, we will just tell a computer to generate one for us.

At that point reality will stop making sense. I will be able to ask for a photo of me, the pope and Fidel Castro riding a pony, and will get a perfectly credible image.

I think that this will change our relationship with photography.